[PPC64LE] Generate correct little-endian code for v16i8 multiply
The existing code in PPCTargetLowering::LowerMUL() for multiplying two
v16i8 values assumes that vector elements are numbered in big-endian
order. For little-endian targets, the vector element numbering is
reversed, but the vmuleub, vmuloub, and vperm instructions still
assume big-endian numbering. To account for this, we must adjust the
permute control vector and reverse the order of the input registers on
the vperm instruction.
The existing test/CodeGen/PowerPC/vec_mul.ll is updated to be executed
on powerpc64 and powerpc64le targets as well as the original powerpc
(32-bit) target.
llvm-svn: 210474
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index d5a685d..91366e2 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -6059,6 +6059,7 @@
LHS, RHS, Zero, DAG, dl);
} else if (Op.getValueType() == MVT::v16i8) {
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
+ bool isLittleEndian = PPCSubTarget.isLittleEndian();
// Multiply the even 8-bit parts, producing 16-bit sums.
SDValue EvenParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuleub,
@@ -6070,13 +6071,24 @@
LHS, RHS, DAG, dl, MVT::v8i16);
OddParts = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, OddParts);
- // Merge the results together.
+ // Merge the results together. Because vmuleub and vmuloub are
+ // instructions with a big-endian bias, we must reverse the
+ // element numbering and reverse the meaning of "odd" and "even"
+ // when generating little endian code.
int Ops[16];
for (unsigned i = 0; i != 8; ++i) {
- Ops[i*2 ] = 2*i+1;
- Ops[i*2+1] = 2*i+1+16;
+ if (isLittleEndian) {
+ Ops[i*2 ] = 2*i;
+ Ops[i*2+1] = 2*i+16;
+ } else {
+ Ops[i*2 ] = 2*i+1;
+ Ops[i*2+1] = 2*i+1+16;
+ }
}
- return DAG.getVectorShuffle(MVT::v16i8, dl, EvenParts, OddParts, Ops);
+ if (isLittleEndian)
+ return DAG.getVectorShuffle(MVT::v16i8, dl, OddParts, EvenParts, Ops);
+ else
+ return DAG.getVectorShuffle(MVT::v16i8, dl, EvenParts, OddParts, Ops);
} else {
llvm_unreachable("Unknown mul to lower!");
}